Last update: January 2022

4 mins to read - 2022/01/07

Next-Gen Personalization With Machine Learning and Device-Side Notifications

Personalization is the golden rule when it comes to marketing. And especially when it comes to mobile engagement, which stands poised to take off as the marketing industry’s #1 channel for brands connecting with users, personalization can make or break a push notification campaign. Personalized content and a moment of delivery customized to your app user’s exact preferences is what distinguishes a sophisticated, high-quality push campaign with a high click-through rate… as opposed to spam. But to achieve truly next-gen personalization that sparks your users’ attention and gets results, machine learning plus device-side computing is the key that will make your push campaign fly.

Download our Metrics & Measurement Playbook to learn how to optimize your push campaign!

personalization machine learning algorithm

How Machine Learning Enables Personalization

This late into the game, machine learning (ML) is pretty standard in mobile engagement SDKs – or, indeed, any platform that leverages user data for whatever reason. ML algorithms are responsible for steering our lives in many ways. For example, algorithms build patterns based on what kinds of videos we watch on Netflix, or listen to on Spotify, to direct us to new music or series that it predicts will be interesting to us. Algorithms can analyze our purchasing patterns to gauge what kind of items and styles we most enjoy, and when it might be most beneficial for us to hear about new products.

Essentially, algorithms have a wealth of user data to calibrate themselves by, and the more data they have to work with the more accurate the algorithms become. Once you get your data set you want to work with, you train your algorithm on that set, and then apply it to specific use cases.

Plenty of research has been done on the effects personalization has on goal conversions. And there’s nothing underhanded or tricky about it; most users enjoy and have even come to expect a personalized marketing experience. Research by Statista has shown that 90% of consumers find marketing personalization somewhat or very appealing. And it makes sense. If you’re going to be viewing ads or reading notifications, you’re going to want them to be catered towards your preferences, rather than something you have no interest in.

However, while other marketing channels have mostly absorbed this wisdom, true personalization – the kind provided by machine learning and AI – is something the mobile engagement industry still struggles with.

Machine Learning + Device-Side Processing = Holy Grail of Personalization

When it comes to mobile engagement, personalizing the content of your push notifications is just the beginning. Most of us take our mobile devices with us wherever we go, and whether we have the attention to devote to reading a notification can change from moment to moment. This is why being able to time your notifications so that they deliver reliably at the right moment for your user to interact with it is the missing puzzle piece that the push notification industry needs.

reliability device-side computing

OpenBack is the first mobile engagement platform to ensure reliable deliverability of its notifications to the device. It achieves this by leveraging machine learning plus device-side computing, whereas other SDKs use the old structure of processing user data on the cloud. With OpenBack, user data never has to leave the device, meaning the entire process is more streamlined, and doesn’t incur the time lag and occasional jumbled or even lost notifications that comes with relying on a 3rd-party for delivery.

This means that for the first time ever, push notifications can apply adaptive scheduling to their push campaigns. And OpenBack has built 40+ custom data signals into its software, enabling mobile marketers to have full control over what moment their notification delivers.

For example, while we do have the “Now” and “Time” signals, which deliver a notification instantly or during a set window of time respectively, there are many more interesting signals to use in combination. For example, there are a series of signals that depend on a user’s reachability. There’s no point in sending a notification at a moment when a user is unable to receive it, such as when their Wi-Fi or data is switched off, or their device is in airplane mode. Once their device becomes reachable again, your notification will be just one among many that swarm them. And they will likely swipe them all away at once.

OpenBack offers different signals that can gauge whether the user is capable of receiving notifications. Or even of whether they are likely to notice your notification, such as our signal that tells you their battery or volume level, their device’s physical orientation (whether it’s faced up or down), whether their lockscreen or headphone jack is engaged, and much more.

Device-Decisions Machine Learning (DD-ML) Signal for Ultimate Delivery Personalization

OpenBack’s most sophisticated signal is our Device-Decisions Machine Learning signal, or DD-ML. It does exactly what the name says: it trains a machine-learning algorithm on a data set of a user’s on-device behavior to create a pattern of their device-usage habits. It then uses real-time contextual factors to moderate that pattern as changing situations may dictate.

For example, if data shows zero device usage from 11 pm until 7 am, followed by a quick burst of use, and then very infrequent device use until 1 pm, the algorithm takes that into account. This particular user likely works a 9-to-5 job, and does not use their phone until their lunch break at 1. Therefore, a mobile marketer could ascertain with the algorithm’s help that the best time to send a push notification offering a deal for a new Subway restaurant near the app user’s office is at 12:55 pm, when they’re at their hungriest and right before they make a decision of where to go for lunch.

However, suppose the pattern changes one day. On Thursday, the user wakes up at 6 am, and their entire device usage pattern shifts an hour earlier. (Perhaps they want to leave early, or perhaps they want to get some extra work done.) The DD-ML signal would take this into account, and would send the notification at 11:55 pm, assuming based on real-time context that the user will be taking their lunch break an hour early.

Early studies on the DD-ML signal have shown using it results in 39% click-through rates for notifications. And the potential for what it could achieve is very exciting for mobile marketers.

To learn more about use cases for the Device-Decisions Machine Learning signal, and other ways to boost personalization in your push campaign, get in touch with one of our experts.

Leave a Reply

Your email address will not be published.

3 × 1 =

Download our Mobile Marketing Playbook to perfect your user engagement game!


Translate »